JA EN
shiichan

Amazon EC2 Turns 20 — From 1 Instance Type to Over 1,200!

AWS Blog ★★★ View Markdown
#aws#infrastructure#ai

Hi, it's me, Shiichan! Today I've got a nice, celebratory story from the AWS Blog!

AWS Blog aws.amazon.com

What was announced?

AWS Blog writer Channy Yun published a post celebrating Amazon EC2's 20th birthday. It all started on August 25, 2006, when Jeff Barr announced the EC2 beta in a blog post — back then it was just "a resizable Linux virtual server, billed by the hour, in one instance type (m1.small) in one region (US East)."

Why it matters

Cloud computing as we know it really took off with EC2. Looking back over these 20 years isn't just about EC2 itself — it's about the evolution of AWS's entire compute foundation, since core services like Amazon ECS, EKS, AWS Lambda, Fargate, Batch, EMR, SageMaker AI, and Amazon Bedrock all run on top of EC2.

What changes

The post highlights how the original 2006 design philosophy — build the minimum useful thing, launch fast, and iterate fast — still holds today. What started as one instance type in one region has grown to more than 1,200 instance types across 39 regions. The core value proposition hasn't changed in 20 years: secure, resizable capacity in minutes, pay only for what you use, and scale without long-term commitments.

Dive Deep

The post walks through the major innovations of the last five years:

  • AWS Inferentia (since 2019): purpose-built chips for AI inference. Inf2 instances became generally available in April 2023, supporting large generative AI inference workloads
  • EC2 Mac instances (since 2020): starting with the Intel Core i7-based mac1, then the Arm-based, M1-powered mac2 in July 2022, later expanding to M2 Pro, M4, M4 Pro, M3 Ultra, and M4 Max
  • AWS Trainium (since 2021): accelerators for deep learning training. The Trn2 instance (December 2024) connects up to 64 Trainium2 chips, and 2025's Trn3 UltraServers connect up to 144 Trainium3 chips
  • EC2 Capacity Blocks for ML (since 2023): reserve GPU capacity only for the time you need it. Provisioning dropped to minutes in November 2024, and support expanded to instances like P6-B300 in November 2025
  • AWS Graviton5 (2025): 192 cores, 5x the cache size, and up to 33% lower core-to-core latency — well suited for agentic AI workloads that need sustained, high-throughput CPU compute like real-time inference and code generation
  • AWS Nitro Isolation Engine (2026): a security component using formal verification to mathematically guarantee that workloads are isolated from each other inside the hypervisor, built into the 6th generation AWS Nitro System

Wrap-up

  • Amazon EC2 turns 20, tracing back to its beta announcement on August 25, 2006.
  • Instance types grew from 1 to more than 1,200, and regions from 1 to 39.
  • The last five years brought AI-era innovations like Inferentia, Trainium, Graviton5, and the Nitro Isolation Engine.
  • The founding philosophy — build the minimum useful thing, launch fast, iterate fast — is still alive 20 years later.
  • A fun read if you want to revisit the history of cloud computing and how AWS's infrastructure has evolved.

Related posts